A core trait for shallow/pointer equality

[bstrie]

< Dzmitry> I have 2 pointers to a shared structure (@). How do I know it's the
           same structure?
< Dzmitry> comparing this pointers issues a compile error saying that @Eq is not
           implemented for the structure
<@pcwalton> Dzmitry: box::ptr_eq
<@pcwalton> Eq is deep
<@pcwalton> there is a separate method for it
<@pcwalton> "it" == shallow comparison
< Dzmitry> pcwalton: interesting. But what if my pointers are wrapped in
           Option<>, can I compare two options by ptr_eq?
<@pcwalton> no, you can't at the moment
<@pcwalton> probably PtrEq or ShallowEq should be a trait like Eq
<@pcwalton> and then implemented on Option<T>

[pnkfelix]

I don't think this is a backwards-compatibility issue. (We should be able to add a ShallowEq trait later.)

I do agree that we probably want this.

Re-nominating for milestone #5: Production Ready.

[glaebhoerl]

FWIW is ptr_eq safe in the presence of potentially moving GC? In Haskell they call it reallyUnsafePtrEquality# for that reason among others, and while Rust's functions aren't required to be pure (so it wouldn't be unsafe in the Rust sense), do similar concerns apply?

[pnkfelix]

@glehel A safe pointer-equality operation can definitely be implemented in a moving GC. One can look for example at how the eq? operation is implemented in Scheme runtime systems.

The place where I would anticipate issues with a moving GC would be unsafe casts from pointers to integers, as one would do e.g. for a pointer-based hashcode operation. So if one wanted a hashtable based on memory address that could use @ptrs as keys, then that hashtable implementation would need to be tied to the moving GC implementation. (See e.g. Larceny's hashtable implementation for an example; that is an example of language implementation that takes pains to not allow safe code to extract identify-based hash codes from arbitrary objects, the way that one can in Java via System.identityHashCode, though some Scheme systems do not take similar pains)

[graydon]

accepted for feature-complete milestone

[pnkfelix]

assigning P-low.

[thestinger]

The ref_eq/ptr_eq methods for references and Rc were completely removed... I'm not sure if that decision was a good one, but it doesn't make sense to add a trait if even a standalone method isn't wanted.

[thestinger]

Rust moved away from providing these functions at all, so a trait doesn't make sense anymore.